Meta-analysis examining fetal sex-specific placental DNA methylation intensities and estimated cell composition post IVF

Abstract

Infertility impacts up to 17.5% of reproductive-aged couples worldwide. To aid in conception, many couples turn to ART, such as IVF. IVF can introduce both physical and environmental stressors that may alter DNA methylation regulation, an important and dynamic process during early fetal development. This meta-analysis aims to assess the differences in the placental DNA methylome between spontaneous and IVF pregnancies. Potential datasets were identified by searching the NCBI Gene Expression Omnibus (GEO) using keywords related to IVF in human participant studies published before November 2023. In our combined fetal sex population (N = 575) from three eligible GEO datasets, 127 autosomal cytosine guanine dinucleotides (CpGs) were significant (False Discovery Rate (FDR) <0.05) between IVF (n = 96) and spontaneous (n = 479) placentae, with 47 CpGs considered differentially methylated (FDR < 0.05 and |Δβ| > 0.05). Stratification by fetal sex revealed no significant autosomal CpGs in fetal female placentae (N = 281); however, in the fetal male placentae (N = 294), we identified nine autosomal CpGs that reached statistical significance between IVF (n = 56) and spontaneous (n = 238) placentae, with three CpGs considered differentially methylated. Fetal male placentae had lower proportions of trophoblasts (P < 0.0001) and stromal cells (P = 0.007) and higher proportions of syncytiotrophoblasts (P = 0.0001) compared to fetal female placentae, regardless of conception type. IVF placentae had higher proportions of stromal cells (P = 0.01) and lower proportions of syncytiotrophoblasts (P = 0.01) compared to spontaneous placentae, regardless of sex. Controlling for cell-type proportions in linear models reduced test statistic inflation and identified new significant CpGs that may previously have been masked by cell-type heterogeneity. The results of this meta-analysis are critical to further understand the impact of IVF on tissue epigenetics, which may help with understanding the connections between IVF and negative pregnancy outcomes. Additionally, our study suggests that sex-specific differences in placental DNA methylation and cell composition should be considered as factors for future placental DNA methylation analyses.

Introduction

In the last 30 years, infertility rates in reproductive-aged couples have increased from 8 to 12% up to roughly 17.5% (Ombelet et al., 2008; World Health Organization, 2023). Rising infertility rates are partly linked to trends toward later motherhood in many developed countries, driven by higher levels of maternal education, increased female employment, and financial uncertainty (Sobotka and Beaujouan, 2018; Molina-García et al., 2019). ART encompasses many medical resources that aid in conception for those struggling with infertility, including those of delayed reproductive age, couples in same-sex relationships, those wishing to use monogenic disease screening, and those opting for frozen or donor oocytes. With a growing demand for ART, investigations of the impact of ART techniques on maternal and fetal health is crucial to ensure a standard level of care for all those in need of reproductive assistance.

ART procedures can include in vivo techniques such as IUI or IVF techniques, including ICSI (Cunningham, 2017). Currently, IVF is the most commonly used ART (Jain and Singh, 2022; Mardovich et al., 2023), with over 8 million children born from IVF in the last 40 years (Fauser, 2019). In the process of IVF, oocytes are retrieved, typically following a period of ovarian stimulation with the use of exogenous gonadotropins and are fertilized by donor sperm in media that mimic the in vivo reproductive environment (Simopoulou et al., 2018; Jain and Singh, 2022). Embryos are incubated until the cleavage stage (Day 3) or blastocyst stage (Day 5), with the latter being more common, before transfer to the uterus (Jain and Singh, 2022).

Despite the benefits of IVF in aiding conception, the procedure introduces additional stressors to early development that are not observed in spontaneously conceived pregnancies. Ovarian stimulation and subsequent oocyte retrieval contribute both hormonal and physical disturbances (Jwa et al., 2019; Sciorio and El Hajj, 2022). Embryo incubation conditions, including oxygen concentration, pH, temperature, and osmolarity, are highly dynamic within the reproductive tract and difficult to replicate in an in vitro setting. Estimated measurements of these biological conditions are mainly from animal studies, which may not be optimal for human development (Ng et al., 2018; Simopoulou et al., 2018). Other factors, such as fresh or frozen embryo transfer and use of preimplantation genetic testing (PGT), may further contribute to the diversity of stressors that IVF may impose on early development.

Current literature has connected ART techniques, including IVF, with an increased risk of many negative pregnancy outcomes. Rates of preterm birth, major birth defects, preeclampsia, cardiovascular issues, and imprinting disorders such as Beckwith–Wiedemann syndrome have all been reported at higher proportions in ART conceived pregnancies (Owen and Segars, 2009; Chen and Heilbronn, 2017; Berntsen et al., 2019; Chih et al., 2021). However, inter-study results in the literature are contradictory (Berntsen et al. 2019).

It has been further hypothesized that epigenetic factors may play a role in this observed relationship between IVF and an increased risk of negative pregnancy outcomes. Epigenetic modifications can be heritable and can alter gene expression without changing the DNA sequence, through processes such as histone modifications, non-coding RNA interference, and DNA methylation (Gibney and Nolan, 2010; Moore et al., 2013). DNA methylation occurs through the addition of a methyl group to the fifth carbon of a cytosine that precedes a guanine, typically referred to as a cytosine guanine dinucleotide (CpG) (Jones, 2012; Moore et al., 2013). This mechanism is important for regulating transcription in processes such as X-inactivation in XX individuals, cell differentiation, and genomic imprinting (Kessler et al., 2018). Both the establishment of DNA methylation and de-methylation can occur dynamically, providing a level of plasticity that may be influenced by environmental factors. While epigenetic modifications can facilitate gene expression changes in response to environmental factors (Li et al. 2021), the extent and mechanism by which DNA methylation contributes to the interaction between the environment and gene regulation remains poorly understood (Law and Holland, 2019).

Dynamic epigenetic regulation is an important process in early embryo development. Following fertilization, there is a genome-wide loss of DNA methylation, aside from imprinted regions, prior to totipotency acquisition and cell differentiation (Barberet et al., 2022; Reyes Palomares and Rodriguez-Wallberg, 2022). Once the blastocyst stage is reached, re-methylation occurs with distinct and asynchronous patterns in the trophectoderm, which will differentiate into the main cell types of the placenta and primitive endoderm (Barberet et al., 2022). As the main steps of IVF occur during this window of DNA methylation reprogramming, it is questioned whether exposure to in vitro conditions may negatively impact this process. Therefore, the connection between IVF stressors and an increased risk of negative pregnancy outcomes may be due to the impact of IVF on DNA methylation in early development, which may lead to gene dysregulation in cell types crucial for placental health.

Assessment of placental DNA methylation patterns in IVF pregnancies compared to spontaneous pregnancies may give insight into the effects of IVF on placental development. Recent meta-analyses have qualitatively compared the results of studies that analyzed DNA methylation data from ART pregnancies in various tissues (Barberet et al., 2022; Schaub et al., 2024). One limitation of these meta-analyses is that many of the included studies used fundamentally different techniques for normalization and statistical analyses in their investigation of DNA methylation. In our study, we aimed to combine and re-evaluate raw, publicly available DNA methylation datasets from placental tissue following IVF or spontaneous conception in a quantitative meta-analysis. We hypothesized that placental DNA methylation levels would be different between IVF and spontaneously conceived pregnancies.

Methods

Dataset collection

This study adheres to the guidelines outlined in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement (Page et al., 2021). Institutional review board ethics approval was not needed for this meta-analysis, as all data obtained were publicly available. Potential datasets were identified by searching the NCBI Gene Expression Omnibus (GEO) using keywords related to IVF in human participant studies published before November 2023. Datasets were first selected on the basis of their titles and abstracts. Datasets included after the first round screening underwent full paper review. Studies were included if they measured DNA methylation from both IVF and spontaneously conceived healthy, full-term, singleton pregnancies. Full-term placenta samples were selected to ensure tissue sampling, and clinical data collection did not occur prior to development of pregnancy complications that may introduce pathology-specific placental DNA methylation patterns. Preterm placentae were excluded as they are known to have different placental DNA methylation profiles compared to full-term controls (Schuster et al., 2019). Studies were excluded if they did not report both IVF and spontaneous samples, were from pregnancies with only multiples, or only had samples from complicated or premature pregnancies. A flowchart showing inclusion and exclusion screening decisions is shown in Fig. 1. All screening steps were performed by one researcher (M.L.). A detailed document was maintained listing all inclusion and exclusion decisions to reduce bias and allow reproducibility.

Figure 1.

PRISMA flowchart for study screening. Flowchart of dataset selection from Gene Expression Omnibus (GEO) based on inclusion and exclusion criteria, assembled using the PRISMA 2020 guidelines.

Screening NCBI GEO revealed three studies that had raw DNA methylation files measured by either Illumina Infinium 450K or EPIC bead chips in placental tissue collected at time of delivery from singleton, uncomplicated, full-term pregnancies (GSE120250 (Choufani et al., 2018), GSE75248 (Paquette et al., 2016), GSE208529 (Huang and Teh, 2022)). Illumina Infinium DNA methylation arrays rely on probe technology, wherein each probe of the microarray binds to specific CpGs and measures single-base resolution of DNA methylation intensities using fluorescence across promoters, open sea methylation sites, gene bodies, and untranslated regions (Zhang et al., 2012; Pidsley et al., 2016). Each study reported conception type, fetal sex, sentrix position, and sentrix ID, and raw intensity data (IDAT) files were available for each sample. Placental collection methods were comparable between studies, with each taking sample biopsies from the fetal side of the placenta directly after delivery.

Data quality control and normalization

R version 4.4.1 (R Core Team, 2023) was used for all subsequent data processing and analysis. All clinical and technical information reported for each study sample was extracted using the GEOquery package (Davis and Meltzer, 2007) version 2.72.0.

IDAT files containing intensities for each probeset on Illumina 450K and EPIC arrays were downloaded from NCBI GEO for each included dataset onto a server. Samples were removed from the datasets if they used in vivo techniques (n = 21) such as IUI, were maternal-facing placental samples (n = 195), did not report conception type (n = 9), or were specified as outliers (n = 9). Meta-data for fetal sex and conception type were downloaded for each sample in each study. As this is public data, we added an additional check step to compare the fetal sex reported in the study meta-data to an estimated fetal sex prediction. Fetal sex was predicted based on DNA methylation intensities from the XY chromosomes using the minfi package (Aryee et al., 2014) version 1.50.0. Only 5 of our 580 remaining samples (0.86%) had mismatched reported and predicted fetal sex, which supports that our fetal sex predictions are valid and that these five samples likely were mislabeled in the study meta-data. These samples were excluded from our dataset, as it could not be confirmed if other meta-data, such as conception type, were also mislabeled. This left a combined sample size of 575 for analysis (n = 96 IVF, n = 479 spontaneous), as shown in Table 1. Background correction and normalization of DNA methylation intensities were implemented using adjusted Funnorm (Wang et al., 2022) through the wateRmelon package version 2.11.2.

Table 1.

Sample demographics for each of the three GEO studies used in this meta-analysis.

	GSE120250	GSE75248	GSE208529	Total
N =	58	321	196	575
IVF sex (M:F)	10:8	6:6	40:26	56:40
Spontaneous sex (M:F)	21:19	153:156	64:66	238:241

GEO, Gene Expression Omnibus; M: male; F: female.

Probe filtering

As GSE208529 had EPIC array data, we first filtered this dataset for 450K array probes to allow comparison to GSE120250 and GSE75248. Additional probe filtering was completed on X and Y chromosome and autosome probes using a modified DNA methylation processing pipeline to allow integration of X and Y chromosome data, as outlined by Inkster et al. (2023). Bad-quality probes, defined as having a missing β value <0.05 or detection P-value <0.05 in >5% of samples, were removed (n = 2189). Probes that bind to known Single Nucleotide Polymorphism (SNP) regions (n = 19361) or that are known to cross-hybridize to multiple sites (n = 36571) were removed as per the annotation from Price et al. (2013). SNP and cross-hybridizing probes are routinely excluded due to the probe-based protocol of an Illumina array. SNPs may impair probe binding based on known genetic variability between samples in these probe regions. These probes may skew recorded methylation intensities in these regions based on genetic variation, as opposed to true differences in DNA methylation status (Daca-Roszak et al., 2015), and were therefore excluded. We filtered out non-variable placental CpG probes (n = 89 160) based on the annotation from Edgar et al. (2017). This step was completed to increase statistical power by reducing the number of multiple tests in CpGs that are specifically non-variable in the placenta (Edgar et al., 2017). After probe filtering, we analyzed a total of 296 545 autosomal CpGs and 8779 X chromosome CpGs. An additional 227 Y chromosome CpGs were analyzed in fetal male samples.

Differential DNA methylation analysis

DNA methylation intensities measured at individual probe-specific CpGs were compared between IVF and spontaneous placental populations using linear regression models implemented through the limma package version 3.60.4 (Ritchie et al., 2015). All linear modeling was conducted using M-values, as they better satisfy the model assumptions (e.g. homoscedasticity and normality of residuals) compared to beta values. Empirical Bayes moderation was applied to stabilize probe-wise variance estimates by shrinking them toward a global prior, improving inference, particularly in small sample contexts. This was first conducted in the mixed fetal sex population using a model that corrected for conception type, GEO study number, and fetal sex.

$$y=\text{Bx}\left(\text{ConceptionType}\right)+\text{Bx}\left(\text{Study}\right)+\text{Bx}\left(\text{FetalSex}\right)+ϵ$$

Following this, fetal sex-stratified populations were analyzed using a model that corrected for conception type and GEO study number only.

$$y=\text{Bx}\left(\text{ConceptionType}\right)+\text{Bx}\left(\text{Study}\right)+ϵ$$

GEO study numbers, which connect samples to each of the three datasets, were used to mitigate batch effects that may be introduced when cross-analyzing data from separate datasets together. We opted not to use ComBat, a common batch effect correction method, as previous studies reported that ComBat may increase false positives in unbalanced study designs and decrease reproducibility of findings (Buhule et al., 2014; Zindler et al., 2020). As this meta-analysis compared studies that were not designed to balance each other, we wanted to limit the potential of ComBat introducing spurious findings, indicative of applying it to an unbalanced study design.

As a sensitivity analysis, we re-ran: (i) a robust variance-corrected option in addition to the original model to further mitigate the influence of outlier variance, often due to small or imbalanced samples, and (ii) the same analysis but excluded samples from GSE75248, which drove the imbalance between IVF and spontaneous samples. Results of this sensitivity analysis were compared to those from our initial model to evaluate any effects of sample imbalance on our statistical findings.

Following linear modeling, statistically significant CpGs were defined as having a False Discovery Rate (FDR) <0.05. Differentially methylated CpGs were defined as having a FDR <0.05 and a change in DNA methylation (|Δβ|) >0.05. Δβs were calculated for each probe by subtracting the mean DNA methylation β value in the spontaneous population from the mean β value in the respective IVF population. Differentially methylated CpGs reaching the more stringent criteria are more likely to indicate a biologically meaningful result.

Placental cell deconvolution

Placental cell-type proportions were estimated using the planet package (Yuan et al. 2021) version 1.12.0 and EpiDISH package version 2.20.0 using placental reference CpGs (Yuan et al. 2021). Proportions of trophoblasts, stromal cells, Hofbauer cells, endothelial cells, nucleated red blood cells, and syncytiotrophoblasts were calculated for each sample. This placental cell deconvolution method is accomplished under the assumption that different placental cell types display distinct DNA methylation signatures, with the whole placental sample reflecting respective frequencies of each placental cell type (Yuan et al., 2021). Therefore, comparing each bulk placenta tissue sample to placental cell-type reference DNA methylomes allows the estimation of cell-type proportions within each sample. Placental cell-type proportion estimations were compared between fetal sex (female versus male) and between conception type (IVF versus spontaneous) using a mixed effects model with study (GSE number) as a random effect using the R package lme4 version 1.1.35.4 (Bates et al., 2015). To investigate possible interactions, we compared sex-specific cell-type proportion stratified by conception type and then cell-type proportion in IVF and spontaneous placentae in samples stratified by fetal sex.

Proportions of cell types were also incorporated as fixed effects into new linear models of autosomal probes. To assess the variability of probes in cell-adjusted and original models, we calculated inflation factors (λ) by comparing the median of the observed chi-squared statistics, a value derived from comparing raw P-values to the expected median under the null. Elevated λ values may reflect an enriched association signal or unmodeled confounding. QQ plots were produced using the qqman package version 0.1.9 (Turner, 2018).

Results

Combined fetal sex and fetal male-stratified populations revealed differentially methylated autosomal CpGs between IVF and spontaneous placentae

We first assessed placental DNA methylation at individual autosomal CpGs between IVF and spontaneously conceived pregnancies. Statistically significant CpGs were defined as having an FDR <0.05. Differentially methylated CpGs were defined as having an FDR <0.05 and a change in DNA methylation (|Δβ|) >0.05 to denote potential biological significance.

In our combined fetal sex population, we identified 127 significant CpGs between IVF (n = 96) and spontaneous (n = 479) placentae, as shown in Fig. 2A. Of these, 47 CpGs met our threshold of differential methylation (8 CpGs Δβ >0.05, 39 CpGs Δβ <−0.05). Sensitivity analyses confirmed our findings were robust to the sample sizes of the IVF and spontaneous placental groups. A comparison of FDR values for the top 15 significant autosomal probes across the original, robust variance, and study exclusion models can be found in Supplementary Table S1. Following the combined analysis, we stratified our populations by fetal sex. Sex-stratified analysis of DNA methylation in the fetal female population revealed no significant or differentially methylated CpGs between IVF (n = 40) and spontaneous (n = 241) placentae, as shown in Fig. 2B. However, in the male-stratified population, we identified nine CpGs that reached statistical significance between IVF (n = 56) and spontaneous (n = 238) placentae, as shown in Fig. 2C. Seven CpGs were hypomethylated in the IVF compared to spontaneous placentae, with LIPJ, C19orf25, MUM1, RASGRF1, FAF1, AK092451, and EEF1A2 being the closest transcriptional start sites. Two CpGs were hypermethylated in the IVF compared to spontaneous placentae, with their closest transcriptional start sites being EXD3 and FBRSL1. The CpGs closest to LIPJ, EEF1A2, and FBRSL1 also met our criteria to be classified as differentially methylated CpGs. These three differentially methylated CpGs were significant in the combined fetal sex population as well.

Figure 2.

Autosomal CpG methylation compared between IVF and spontaneous placentae. Volcano plots depicting differentially methylated autosomal sites between IVF and spontaneous placentae in (A) combined fetal sex population, (B) female-stratified population, and (C) male-stratified population. –log10 of the adjusted P-value, including indication of False Discovery Rate (FDR), is plotted on the y-axis, and change in DNA methylation (Δβ) is plotted on the x-axis. CpGs highlighted in blue are differentially hypomethylated in IVF (Δβ <−0.05, FDR <0.05). CpGs highlighted in yellow are differentially hypermethylated in IVF (Δβ >0.05, FDR <0.05). Sample size is indicated under the x-axis.

No differentially methylated CpGs were identified in sex-stratified analyses of XY chromosomes

We assessed DNA methylation in the X chromosome between IVF and spontaneous placentae in sex-stratified populations. In the fetal female population, we observed no significant or differentially methylated CpGs between IVF and spontaneous placentae, as shown in Fig. 3A. In the fetal male population, we found one statistically significant CpG with the closest transcriptional start site to CAPN6, as shown in Fig. 3B. However, this CpG did not reach our criteria to be considered differentially methylated. Significant findings from autosomal and X chromosome linear models can be found in Supplementary Table S2.

Figure 3.

X-chromosome CpG methylation compared between IVF and spontaneous placentae. Volcano plots depicting X chromosome methylation differences between IVF and spontaneous placentae in (A) female-stratified population and (B) male-stratified population following linear modeling. –log10 of the adjusted P-value, including indication of False Discovery Rate (FDR) is plotted on the y-axis, and change in DNA methylation (Δβ) is plotted on the x-axis.

We assessed DNA methylation in the Y chromosome between IVF and spontaneous placentae in the fetal male-stratified population. No significant or differentially methylated CpGs were observed (Supplementary Fig. S1).

Placental cell-type proportions differed between sex and conception type

We estimated placental cell-type composition in each of our samples using a placental cell deconvolution method. Cell proportions were first compared using a mixed effects model. Between fetal sexes, we observed that fetal male placentae have lower proportions of trophoblasts (P < 0.0001) and stromal cells (P = 0.007), but higher proportions of syncytiotrophoblast (P = 0.0001), compared to fetal female placentae. Furthermore, between conception types, we observed that IVF placentae have higher proportions of stromal cells (P = 0.01) and lower proportions of syncytiotrophoblasts (P = 0.01) compared to spontaneous placentae (Fig. 4). When replicating these models in conception-stratified or fetal sex-stratified populations, the results were reproduced and showed that the effects of sex and conception type on placental cell proportion are largely independent and not confounded.

Figure 4.

Placental cell-type proportions across fetal sex and conception strata. Box plot of proportions of estimated cell types across spontaneous male, spontaneous female, IVF male, and IVF female placental sample strata.

Cell-type adjusted linear modeling of autosomes revealed new significant CpGs in all population strata

As we observed detectable differences in cell type between conception and between sex independently, we wanted to investigate how this may impact our statistical analyses. Cell-adjusted linear models compared autosomal CpGs between IVF and spontaneous in the combined fetal sex, fetal male, and fetal female-stratified populations. In the combined fetal sex group, cell-adjusted linear models identified 138 significant CpGs between IVF and spontaneous placentae (90 overlapping with cell-unadjusted model). Of these, 31 CpGs had a Δβ <−0.05 (28 overlapping with cell-unadjusted model) and three had a Δβ > 0.05 (two overlapping with cell-unadjusted model).

In the fetal male strata, cell-adjusted linear models identified 18 significant CpGs between IVF and spontaneous (eight overlapping with the cell-unadjusted model). Of these, two CpGs had a Δβ <−0.05 (both overlapping with the cell-unadjusted model) and three had a Δβ >0.05 (one overlapping with cell-unadjusted model). In the fetal female strata, two CpGs met statistical significance in cell-adjusted models but did not meet biological significance thresholds. Significant results and overlap from our cell-adjusted models in all three strata can be found in Supplementary Table S3.

Fetal females showed greater raw P value inflation scores across autosomes and cell deconvolution-specific probes, which could be reduced after adjusting for cell type

In the cell-unadjusted autosome CpG model, inflation was observed in whole sample (λ = 1.34) and in female (λ = 1.45) but not fetal male (λ = 0.94) placentae. Adjustment for estimated cell-type proportions reduced inflation across groups (Fig. 5). λ values in the adjusted model were 0.99 (whole sample), 0.97 (fetal male), and 1.25 (fetal female). To assess whether cell-type estimation probes contributed to the observed inflation, we repeated this analysis restricted to the 560 probes used for deconvolution. In the unadjusted model, these probes showed marked inflation (λ = 1.99 whole sample; λ = 2.02 females), which was attenuated in the adjusted model (λ = 0.78 overall) (Fig. 6).

Figure 5.

P value inflation in autosomal probes. QQ plots of raw P values of autosomal probes from cell-unadjusted and cell-adjusted linear models that compared IVF to spontaneous in combined fetal sex, fetal female, and fetal male placenta strata.

Figure 6.

P value inflation in cell deconvolution probes. QQ plots of raw P values of cell deconvolution probes from cell-unadjusted and cell-adjusted linear models that compared IVF to spontaneous in combined fetal sex, fetal female, and fetal male placenta strata.

Discussion

The incidence of ART has more than doubled over the last 10 years, with IVF as the most common method (Mardovich et al., 2023). Despite its benefits in aiding conception, IVF has been associated with increased incidence of adverse health outcomes such as major birth defects, preeclampsia, preterm birth, and increased cardiovascular risks (Chen and Heilbronn, 2017; Berntsen et al., 2019; Chih et al., 2021). The IVF protocol typically occurs within the first 5 days following fertilization (Jain and Singh, 2022), which overlaps with the dynamic epigenetic reprogramming necessary in early preimplantation blastocysts (Barberet et al., 2022). Therefore, it has been proposed that the IVF protocol and associated stressors may impact early DNA methylation, predisposing the pregnancy to a higher risk of negative health outcomes. The purpose of this study was to investigate the impact of IVF on the placental DNA methylome of healthy, full-term singleton pregnancies using publicly available DNA methylation datasets. We hypothesized that there would be differences in DNA methylation in IVF placentae compared to spontaneous controls.

In previous investigations of DNA methylation differences between ART and spontaneous pregnancies, a combination of candidate approaches through pyrosequencing or methylation-specific PCR techniques and whole genome arrays such as Illumina have been utilized (Gomes et al., 2009; Rancourt et al., 2012; Mulder et al., 2020; Mani et al., 2022). Recent systematic reviews (Barberet et al., 2022; Schaub et al., 2024) have qualitatively compared the results of studies that analyzed DNA methylation data from ART pregnancies in various tissues; however, this remains a complex challenge due to diversity in normalization and statistical approaches across many of the compared studies. Our meta-analysis re-analyzed previously published datasets within a consistent statistical workflow, allowing for a more direct and quantitative comparison of sample data.

Screening of datasets on NCBI GEO yielded three studies with raw accessible data that met our inclusion criteria for the meta-analysis. Linear modeling techniques were used to compare placental autosomal DNA methylation in IVF and spontaneous placentae in a combined fetal sex population. We identified 127 statistically significant (FDR <0.05) CpGs between IVF and spontaneous placentae in our combined fetal sex population, with 47 CpGs reaching our more stringent threshold to be considered differentially methylated (8 Δβ > 0.05, 39 Δβ <−0.05). Our sensitivity analyses confirmed reliable performance of our linear models, despite differences between IVF and spontaneous sample counts in this analysis. Three significantly and two differentially hypomethylated CpGs mapped to the closest transcriptional start site of IRF7, which encodes an important immune regulator. Higher IRF7 gene expression has been observed in immune cells of preeclamptic mouse models following pro-inflammatory macrophage injection (Fei et al., 2025). Another single CpG with differential hypomethylation had the closest transcriptional start site to TNRC18. Overexpression of circTNRC18 can result in negative trophoblast cell migration and is associated with preeclamptic pregnancies (Shen et al., 2019).

In a recent meta-analysis, Andrews et al. (2022) identified placenta-specific differences in DNA methylation by fetal sex at both site-specific and regional levels, suggesting that analyzing male and female placentae together as a single population may be inappropriate. To address this concern, we stratified our data by fetal sex. In fetal female placentae, there were no significant CpGs between IVF and spontaneous groups. However, in the fetal male-stratified population, we identified nine autosomal CpGs that reached statistical significance. Two hypomethylated CpGs mapped to the closest transcriptional start sites of EEF1A2 and LIPJ, while one hypermethylated CpG mapped to FBRLS1. These three CpGs were also identified in our combined fetal sex analysis. FBRLS1, which encodes FBRSL1, a protein involved in RNA binding (Jedynak et al., 2023), has been found to be differentially methylated in cord blood and blood spots of ART pregnancies (Barberet et al., 2022). LIPJ appears as significantly enriched in gene sets for gestational diabetes (Zhang et al., 2023) and has been previously differentially methylated (Wilson et al., 2018) and differentially expressed (Leavey et al., 2018) in preeclampsia placentae. EEF1A2 has shown novel function in angiogenesis through positive feedback regulation via HIF1A (Patel et al., 2024), an important process in early gestation that is often dysregulated during placental pathologies such as preeclampsia (Sahai et al., 2017). This is of interest, as both gestational diabetes and preeclampsia have been observed at a higher incidence in IVF pregnancies (Chih et al., 2021; Ghanem et al., 2024). Further gene expression analysis may also identify if differential DNA methylation in these regions impacts the enhancers, which may affect nearby genes such as KCNQ2 and RNLS, both previously investigated in the context of preeclampsia (Mistry et al., 2011; Wei et al., 2018; Soliman et al., 2025).

We identified one statistically different CpG on the X chromosome of male placentae in IVF compared to spontaneous samples, which mapped to the closest transcriptional start site of CAPN6. Although this site did not reach our threshold to be considered biologically significant, investigation into this gene in the future may be interesting, as it is highly expressed in placental tissue and may have various roles in pathology and disease (Chen et al., 2020).

An important consideration when interpreting our findings is what may be driving the observed differences in DNA methylation. Previous reports (Andrews et al., 2022; Campbell et al., 2023) have indicated that differences in placental cell proportions may account for differences and heterogeneity in placental DNA methylation and gene expression. In our analysis, we found that both sex and conception type can independently impact placental cell-type proportions. IVF placentae had higher proportions of stromal cells and lower proportions of syncytiotrophoblasts compared to spontaneous placentae. Additionally, fetal male placentae had higher proportions of syncytiotrophoblasts and lower proportions of trophoblasts and stromal cells compared to fetal female placentae, regardless of conception type. This contrasts the findings of Auvinen et al. (2024) that identified higher trophoblasts and lower stromal cell populations in ART versus spontaneous placentae. Furthermore, in their IVF-only samples, only stromal cells were significantly lower compared to controls and likely driven by signals in female placentae (Auvinen et al., 2024). When adjusting for estimated cell composition in our autosome linear model, we see reduced test statistic inflation, specifically in females, compared to cell-unadjusted models. This is even more apparent in the inflation scores of probes specifically used for cell-type deconvolution. This suggests that cell-type adjustments improve specificity and reduce false-positive signals. Without adjustment, we observe an enrichment of variability in fetal female placentae, likely driven by unmodeled cellular heterogeneity. As we also see differences in cell-type proportions between conception types, it may explain why other studies have noted higher variability in placental DNA methylation in IVF groups compared to spontaneously conceived populations (Choux et al., 2018; Mulder et al., 2020).

When we adjusted models for cell type, beyond changes in inflation scores, we also observed differences in the significant autosomal CpGs identified. The appearance of new CpGs (48 in combined fetal sex, 10 in male-stratified, and 2 in female-stratified analyses) in these models may reflect DNA methylation signals that were previously masked by cell-type heterogeneity. In the combined fetal sex population, two additional CpGs were identified near the closest transcriptional start site of COL6A1, with one reaching the threshold for biological hypomethylation. Notably, COL6A1, a major extracellular matrix protein that promotes migration, invasion, and endothelial tube formation, has characterized downregulation in the placentae of preeclamptic humans (He et al., 2015; Qi et al., 2025) and in low-birthweight bovines (Wang et al., 2024). Additionally, a hypermethylated CpG was identified near MUC6, a member of the mucin family that has shown enriched mutations in the placentae of those with recurrent pregnancy loss (Lin et al., 2025). For CpGs that remained significant in both the cell-unadjusted and cell-adjusted models, including the aforementioned CpGs near IRF7, TNRC18, FBRSL1, EEF1A2, and LIPJ, it is possible that the difference between IVF and spontaneous is large and dominated in one cell type, or that the differences in methylation signals occurred similarly across multiple cell types.

Interestingly, the published results of the individual datasets of GSE120250 (Choufani et al., 2019) and GSE75248 (Litzky et al., 2017) did not find any significant hits between ART and spontaneous groups, which may further highlight the advantage of a larger sample size in understanding the impact of IVF on placental DNA methylation. We did not compare our findings to published results from GSE208529, as their published results did not focus on a comparison between IVF and spontaneous groups (Kee et al., 2022). Auvinen et al. (2024) tested a similar hypothesis in placental DNA methylation and gene expression from the placentae of different conception groups using a comparable bioinformatics pipeline. This study identified 6814 significant CpGs between ART (n = 80) and spontaneous (n = 77) placentae and 230 CpGs between IVF specifically (n = 50) and spontaneous (n = 77) placentae. When looking at significant autosomal CpGs, 25 overlapped with the Auvinen et al. ART analysis (7 CpGs from cell-adjusted model only, 2 from cell-unadjusted model only, and 16 present in both cell-adjusted and cell-unadjusted models), and 22 overlapped with the Auvinen et al. IVF analysis (8 from cell-adjusted model only, 14 present in both cell-adjusted and cell-unadjusted models). This overlap further validates that our observed differential DNA methylation between IVF and spontaneous placentae is likely robust and reproducible when approached with similar bioinformatics processing, even across differing populations.

The observed trends of DNA methylation and cell composition differences in placental tissue between males and females further highlight how fetal sex may impact placental development and have clinical implications. The placenta matches in sex and genetic characteristics with the developing fetus (Herrick and Bordoni, 2024). Different DNA methylation patterns are observed between male and female placentae, which may contribute to placental sexual dimorphism (Christians, 2022). Hypotheses have previously reported that males may adopt strategies of increased investment toward fetal growth and higher placental efficiency, while females are more responsive to environmental stressors (Eriksson et al., 2010; Braun et al., 2022). Although there are evident sex differences in placental transcriptomics, there are no indications that support a clear increase in responsiveness to the maternal environment by female fetuses or a dangerous approach utilized by male fetuses (Christians, 2022). Furthermore, measures of placental efficiency are widely debated, and calculations of efficiency using birth weight: body weight ratios, as reported by Eriksson et al. (2010), may result in an oversimplification and invalid representation of the true trends of placental function (Christians et al., 2018). Further studies should aim to investigate the implications of differential methylation and cell composition between fetal male and female placentae to elucidate whether these trends are leading to biologically significant differences in placental function in a sex-specific manner following IVF.

A large hindrance and data bias in this study are due to the lack of robust publicly available data published through open science frameworks such as NCBI GEO with accessibility for re-analysis. It is critical that raw data files with correct formatting are updated frequently to allow for quantitative meta-analyses to validate findings across studies, despite differences in inter-study downstream processing methods. We recommend those wanting to share biological data to follow the guidelines as set out by Wilson et al. (2021). Further limitations to working with public data include the retrospective nature of this study, in which we were unable to assess changes in RNA expression from the placentae included. We also had a lack of appropriate controls to fully understand whether these findings are likely a result of the IVF protocol or underlying factors of infertility. Additionally, important clinical data such as the type of embryo transfer (fresh or frozen) and the use of PGT or ICSI were not specified. This highlights that future clinical work in the field of ART should aim to collect more robust data regarding the type of ART used, the presence of PGT, frozen or fresh embryo transfer, and fertility status.

Regardless of these limitations, this study is the first analysis of public data that these authors are aware of that compares DNA methylation in IVF and spontaneous placentae using a harmonized, reproducible analysis pipeline, facilitating cross-study comparability with systematic evaluation of model inflation and sensitivity. We also incorporated sex-stratified analyses, uncovering sex-specific epigenetic alterations associated with mode of conception, as well as sex-specific placental cell composition differences independent of conception type. Estimation of placental cell type across sex and conception type is often overlooked in prior studies or not included in cellular models; however, our findings indicate this may be a necessary step to account for placental heterogeneity when analyzing DNA methylation, to improve both power and interpretation.

To address confounding variables such as fertility status and type of ART used, future work with a mouse study would be beneficial to elucidate if changes in methylation are correlated with IVF protocol involvement or the presence of underlying issues in fertility. Animal models also allow for methylation analysis across gestation, as early methylation differences may contribute to pathology. Additionally, use of animal or cell models may be beneficial to validate if these CpG-specific changes in methylation or observed variability may equate to biologically significant impacts in placental development.

In conclusion, the results of this study are critical to further understand the impact of IVF on tissue epigenetics, which may help to investigate the connections between IVF and negative pregnancy outcomes. Our study shows that sex-specific differences in placental DNA methylation and cell composition should be considered when analyzing placental data to enhance the computational ability to detect biologically relevant and reproducible findings.

Author Biography

Dr Samantha L. Wilson: Dr Wilson leads the Wilson Pregnancy Lab at McMaster University, where her research investigates the molecular origins of pregnancy complications. Her team uses a multi-omic approach to integrate genomic, epigenomic, and transcriptomic data from cell-free DNA and other blood-based biomarkers to study how the placenta develops and adapts to maternal and fetal demands. By applying machine learning to these complex datasets, the lab uncovers molecular pathways underlying disorders such as preeclampsia and fetal growth restriction. This work advances understanding of placental biology and disease etiology, providing critical insight into the mechanisms that shape pregnancy outcomes.

Supplementary data

Supplementary data are available at Molecular Human Reproduction online.

Data availability

All source code can be found on the Wilson Pregnancy Lab Github (https://github.com/WilsonPregnancyLab/IVF_MetaAnalysis_Repo). All data are already on GEO, and how to extract those datasets is within the source code.

Authors’ roles

M.L., W.Q.D., K.W.S., and S.L.W. all substantially contributed to the study conception and finalized manuscript. M.L. screened for publicly available data and completed analyses with assistance from K.W.S. and consultation from W.Q.D. and S.L.W. M.L. was responsible for drafting this manuscript. All authors contributed to manuscript revisions and approved the final version for submission. All authors agreed to be accountable for the accuracy and integrity of this work.

Funding

M.L. holds a Canada Graduate Scholarship—Doctoral Research Award (Canadian Institute of Health Research), funding Reference Number: 199359. K.W.S. holds a Canada Graduate Scholarship—Masters (Canadian Institute of Health Research).

Conflict of interest

The authors report no conflicts of interest.